On the eve of SpaceX's IPO, Musk disclosed in detail for the first time the plan for an orbital AI data center, positioning it as the company's core growth engine, attempting to break through the limitations of Earth's power supply on the development of the AI industry by transferring massive computing power to low Earth orbit.
In a video released on Monday, Musk unveiled for the first time the design sketches and core technical parameters of the first generation AI satellite "AI1".
Musk has made it clear that establishing data centers in space does not require any "magic" that has not yet been invented, and the technological challenges it faces are even lower than those of existing Starlink businesses. This is not an extremely difficult engineering problem for SpaceX to solve, and they will strive to achieve an annualized deployment rate of 1 gigawatt (GW) of space AI computing power per year by the end of 2027.
This latest statement sends a clear signal to the capital market that SpaceX is attempting to transform its absolute scale advantage in satellite production and launch into a moat for the next generation of AI computing infrastructure.
Despite industry competitors being cautious about the economic feasibility of space computing power, SpaceX has applied to the Federal Communications Commission (FCC) to launch up to one million AI satellites. To cope with the high barriers of computing power and launch costs, SpaceX is working with partners such as Tesla and Intel to develop its own chip factories, and plans to use Starship Heavy Rocket to disrupt existing launch economics, which directly widens the gap in strategic deployment with competitors.
Targeting the bottleneck of computing power and the huge potential market
In the submitted IPO application documents, SpaceX pointed out that the potential AI market, expected to reach up to $26.5 trillion, will be severely constrained by the inability of the Earth to rapidly expand its power generation capacity. Therefore, solar powered orbital AI data centers are seen by Musk and aerospace industry executives as a key technology to meet the growing energy demands of AI companies.
Regarding the construction schedule, Musk has made highly ambitious predictions. He stated that SpaceX will strive to achieve an annualized deployment rate of 1 gigawatt (GW) of space AI computing power per year by the end of 2027, and seek to expand at an order of magnitude rate each year, ultimately reaching a computing power scale of 1 terawatt (TW).
However, he also advised investors to "hold a cautious attitude" towards the aggressive timetable, while the official expectations given in the IPO documents are more robust, which is to gradually promote commercialization starting from 2028.
Unveiling AI1 Satellite: Nvidia Cabinets in Orbit
SpaceX has clarified the true hardware form in response to the misconception that space data centers are "launching ground data centers into space". The core challenge of engineering is not the migration of physical buildings, but the acquisition of electrical energy in a vacuum environment and the efficient radiation of waste heat generated by high-power computing power.
The video showcased for the first time the quantitative indicators of AI1 computing satellites. The peak power consumption of the satellite is 150 kilowatts, and the continuous average calculated power consumption is 120 kilowatts.
Musk stated that this metric accurately matches the operating power envelope of the Nvidia GB300 computer cabinet (containing 72 GPUs) used in ground data centers, equivalent to directly sending an entire Nvidia AI computing module into space.
To meet the extremely high energy consumption and heat dissipation requirements, AI1 satellite has a pair of giant wings with a wingspan of 70 meters. The solar array has a power generation density set at 250 W/m ², and the double-sided heat dissipation plate has a heat dissipation density of 1400 W/m ². It will face the sun in a "blade" posture in orbit to achieve maximum heat dissipation.
Simplifying architecture and reusing technology to build a manufacturing moat
In terms of hardware structure, the engineering design of AI1 satellite is even more streamlined than traditional Starlink satellites.
The existing Starlink satellites require extremely complex giant phased array antennas and parabolic antennas, while AI satellites are essentially more like pure large-scale hardware: they are mainly composed of giant solar arrays, super large heat dissipation plates, and basic laser links, eliminating the need for complex ground communication antennas.
Musk and his engineering team emphasized that the vast majority of AI satellites manufactured directly reuse SpaceX's Starlink V3 satellite platform technology, which has been developed and validated.
This means that SpaceX does not need to make breakthroughs in basic science to directly translate its existing experience in mass production, launch, and operation of satellites. As the IPO approaches, this extremely high technology reuse rate and engineering scalability capability constitute the unique competitive advantage that the company presents to investors.
Ecological collaborative response to cost and delay challenges
SpaceX has provided a clear solution to address concerns about network latency caused by space data centers.
AI satellites will be deployed in low Earth orbit (LEO) at a distance of 600 to 800 kilometers from the ground, with a one-way network delay of only about 3 milliseconds. The satellite will integrate an inter satellite laser link with a bandwidth of up to 1 Terabit per second, and transmit data at high speed through Starlink's existing KA and KU band antenna network, or directly through the satellite ground laser link.
However, there are still differences within the industry regarding commercial feasibility.
Blue Origin, along with Amazon founder Jeff Bezos and researcher Andrew McCalip, have pointed out that expensive AI chips and high launch costs are the current industry obstacles, and the current economic model is not yet reasonable.
To this end, SpaceX is attempting to build a vertically integrated supply chain to break down cost barriers: on the one hand, it relies on Starship Heavy Rocket to significantly reduce launch costs, and on the other hand, through a planned factory called Terafab, it collaborates with partners Tesla and Intel to develop and manufacture its own AI chips. By controlling the launch and underlying computing hardware, SpaceX is accelerating its commercialization of space computing.
